Focus: 64-bit Computing
64-bit Computing re-examined

With 64-bit hardware blossoming, 64-bit computing is not just for scientific applications and deep databases. Let's take a look at the porting issues. by Seamus Phan

When we talk or hear about 64-bit computing, high performance computing (HPC) comes to mind. It is common to apply 64-bit computing to applications such as defense, e-government, global financial management, or even genomics and other bio-simulations.

However, 64-bit computing can now be successfully applied in the management of Web caching and search engines for large service providers and upstream data centers, as well as e-commerce, data warehousing, rules and constraints-based programming, and decision support systems.

The primary reason is simple 64-bit computing allows the real-time loading of data into the vastly expanded memory space, thereby allowing immediate access and computation by users of such systems.

Should you port?
So should we extend 32-bit applications to allow our applications to harness the tremendous leap in memory addressing?

For example, computer-aided design (CAD), business modeling, 3D simulation, and semiconductor design are all applications that often have been designed to work in both 32-bit and 64-bit memory spaces. Under 32-bit application space for the likes of Solaris, such applications can utilize up to the 4 GB limit, or 2 GB for Windows.

For many low-end applications, 4 GB limits are not easily breached and most of these applications will serve the needs of users for years to come. However, in more advanced applications, especially those in defense and other high-end decision support systems, embracing the 16 GB space can result in a quantum leap in performance.

For example, in the area of defense, it could be the difference between eradicating threats or being bombed.

From the technical standpoint, you have to ascertain if your application is managing its own memory management, and whether it is wearing out the 32-bit address space.

In addition, your application will benefit from a memory address space of more than 4 GB, and will require the full 64-bit registers to handle 64-bit integer math faster.

Windows XP has a 64-bit version that will allow the simultaneous execution of 32-bit and 64-bit applications. Sun has Solaris which is 64-bit capable. And if you prefer the portability and cost-effectiveness of open source, NetBSD (www.netbsd.org), a cousin of Unix used in the likes of FreeBSD (a purely 32-bit OS for x86 CPUs) and Mac OS X (64-bit possible), has been ported to embrace the AMD x84-64 platform and the MIPS64 platform, to cater to 64-bit applications. This is on top of more than 40 different hardware architecture by Wasabi Systems (www.wasabisystems.com).

Having determined these needs for your specific application, you can then decide how to migrate to the 64-bit application space.

Porting considerations
For Windows environment, you can choose to port your application in three ways a full port; retaining a 32-bit address space with 64-bit pointers; or retaining a 32-bit address space with 32-bit pointers.

Choosing the 32-bit address space with 32-bit pointers is perhaps the easiest. If your application is not likely to traverse the 2 GB memory address limit for Windows, or 4 GB limit for the likes of Solaris, then you may consider retaining the use of the 32-bit address space with 32-bit pointers.

However, using this rather limiting approach has some concerns you will periodically face pointer truncation warnings, because most of the higher bit pointers will be truncated downward to fit within the 32-bit space. If you are using a 64-bit operating system, then this method may not work well because 64-bit pointers are required to interact with a 64-bit OS.

The next step up is to retain the smaller 32-bit address space but equip your application to cater to 64-bit pointers. This method of partial porting will at least make it more compatible with 64-bit operating systems.

In this scenario, your application does not have the need for a larger memory address space, and pointer truncation warnings can be more safely ignored, as the upper 32-bit pointers will be set at zero.

With 32-bit pointers, data size expansion is a key concern, while it is less of a concern with 64-bit pointers.

Obviously, the right way to port is to port fully to 64-bit memory address space, with 64-bit pointers as well. A full port is understandably the most tedious and complex, but your application will be immediately scalable to take advantage of the massive address space beyond 2 or 4 GB (for Windows and Solaris respectively).

All pointer truncation warnings must be ironed out in a full port, including dynamic link libraries (DLLs) and other third party plug-ins. In a full port, both data and code expansion will not be an issue since both address space and pointers are fully 64-bit capable.

In addition, 32-bit modules and 64-bit modules cannot co-exist in the same process in a full port.

For complex high-end applications, your workstation or server should be equipped with full 64-bit ported applications, rather than allowing a mix of 32-bit and 64-bit applications since they may destabilize each other.

Porting tools
While hardcore coders may consider using a bare-bone compiler like the GCC Compiler for porting, there are commercial tools that aid the process of code porting. A company that provides 64-bit porting tools is Migratec (www.migratec.com), which has the 64express product to port 32-bit applications to both the Itanium and PowerPC/AIX5L hardware architecture.

64express automates the process of porting 32-bit C or C++ source code to 64-bit systems for Windows, Linux and AIX 5L. According to Migratec, programmers can identify source codes to target problems that common compilers such as GCC might miss, and eliminate manual transcription errors. 64express can also help to reduce tests and debugging.

It identifies the API (application programming interface) differences between 32-bit and 64-bit target platforms, so that there will be less compiling errors. An important area is data truncation and extension, where 64express helps to identify. Through the automated process, it helps to automate API substitution, and solve truncation issues with programmer-defined rules and constraints. The decent part about 64express is its preservation of the original source code, so that programmers can always return to the original state without problems.

This is an important step in helping programmers learn about the possibilities of porting, while having the latitude to make some mistakes with full control. And of course, 64express also provides fully documented code changes so that more changes can easily be made by other programmers, hence complying with standard programming practices.

There is definitely a blossoming in 64-bit hardware. While 64-bit software appears not to have caught up yet, many server-based applications have been ported, including Apache.

With a dramatically increased memory address space, it is conceivable that low-level server hardware running consumer-grade 64-bit hardware such as Itanium, will be able to leverage on the increased stability and real-time processing of 64-bit server applications.

For MIS, it may be time to start toying with open source and commercially available porting tools, and determine if custom enterprise applications or even middleware and front-ends, can be converted to provide greater scalability and performance.

Seamus Phan is research director at KnowledgeLabs News Center (www.knowledgelabs.net), an independent technology news bureau and writes for Network Computing-The Asian Edition. He can be reached at seamus@knowledgelabs.net